Tool
Give the batch processor a folder of messy documents and say in your own words what you need from them. It gives back a searchable database, a dated record of the facts with their sources, and a cited report and briefing.
Collage after St Jerome in his Study, Antonello da Messina, c. 1475, National Gallery, London ยท with Old Centaur, Aristeas & Papias of Aphrodisias, Capitoline Museums, and Thomas Baines, The Rapids of the Victoria Falls, Zambezi River
Years of contracts, minutes, spreadsheets, scans and old drafts, many of them saved more than once under different names. A new director, an investor or a buyer needs the story in there, and nobody has weeks to read it.
The batch processor reads all of it. You describe what you need once, in plain words, and that description steers every step: what counts as relevant, which facts to pull out and what the report says. So the same tool works for getting a new director up to speed, preparing for due diligence or understanding a drive you have just inherited.
Fast models do the bulk reading, through batch processing that costs about half the price of asking the same questions one at a time. The result is compact and cited, ready for you or a frontier model to reason over.
Describe what you need. Point it at a folder and say what you want from it.
Scan and convert. On your own computer, it finds every file, sets aside exact duplicates and turns each document into text. Nothing has been sent anywhere yet.
Sort what matters. A fast model reads a short excerpt of each document and decides whether it is relevant, with a one-line reason.
Check the sort. You get a spreadsheet of every verdict. Change any you disagree with, then carry on.
Pull out the facts. Each relevant document is read in full, and every fact is recorded with its date, its source and a quote.
Read the report. A cited report, a short briefing and a database you can search.
Any set of documents and any goal: the words you write steer every stage.
Bulk work runs as batch jobs at about half the usual cost, and you see an estimate before anything is sent.
Sorting reads only an excerpt of each document with a fast model. Only the documents that matter get a full read.
The source folder is only ever read. Nothing is moved, renamed or changed.
Two shareholder registers give two dated entries, not one overwritten one, so you see the current position and the timeline behind it.
Scanned PDFs and photos of documents are sent to the model as they are, so they are read rather than skipped.
Close it at any point and start again later. Add new files next month and only the new ones are processed.
Double-click to start and it walks you through each step, or drive every stage from the command line.
Modules are optional add-ons that build on the documents the batch processor has already read. Three examples:
Sorts your documents into a due-diligence data room in numbered categories, writes a cited report for an investor, and lists the documents an investor will ask for that are missing. Add them later and it shows which gaps are now closed. The checklist adapts to the country you name.
Turns a drive of spreadsheets, documents and presentations into a compact pack a frontier model can reason over in one sitting: a map of the drive, a data dictionary, the processes and timeline it shows, and the questions the data can answer. Large spreadsheets are described by their shape rather than every row, so a huge workbook costs about what a memo does.
Pulls structured records out of your documents, such as company details, contracts, suppliers, customers and compliance dates, in a format ready to load into your own workspace. Anything that does not fit the format is left out, never guessed.
Download it from GitHub, where the setup guide takes you through each step. If you would rather we ran it for you, or built a module for your own documents, book a call.